Dynamic process monitoring based on orthogonal dynamic inner neighborhood preserving embedding model

Abstract

A novel dynamic data modeling algorithm named orthogonal dynamic inner neighborhood preserving embedding (ODiNPE) is proposed for dynamic process monitoring. The formulation of the ODiNPE algorithm attempts to optimize a dual objective, which integrates together the maximization of the auto-covariance of the latent factors and the minimization of reconstruction error from the neighborhood, while an orthogonal constraint on the projecting directions is also satisfied. Therefore, the proposed algorithm is expected to extract highly auto-correlated dynamic latent factors with intrinsic neighborhood information embedded. The application of the ODiNPE in dynamic process monitoring has demonstrated its effectiveness and superiority over other state-of-art dynamic process monitoring approaches.